Run test_count_call_alleles on Cubed

[tomwhite]

See #908

[tomwhite]

This is passing all tests, but when producing a visualization of the Dask graph I noticed that it reintroduces the problem in #871, so shouldn't be merged. (cc @timothymillar)

[tomwhite]

In the meeting @jeromekelleher suggested that we could abstract the map_blocks bit into a function in sgkit that dispatches either to da.map_blocks (for Dask arrays) or xr.apply_ufunc (for everything else, including Cubed). This would keep the Dask path as it is today, while allowing other array libraries to work.

[mergify]

This PR has conflicts, @tomwhite please rebase and push updated version 🙏

[tomwhite]

Superseded by #1254